Selective activation of hindered meta-amp;sp2 hydrocarbon bonds facilitated by automation and deep learning

On September 19, 2022, the Liao Minbiao Research Group of Guangzhou Laboratory, in collaboration with yang Yuedong Research Group of Sun Yat-sen University, published a research paper entitled “Selective Functionalization of Hindered meta-C–H bond of o-Alkylaryl Ketones Promoted by Automation and Deep Learning” in the journal Chem.

The results involve CO2 activation, carboxyl-oriented Pd catalytic C–H bond activation, automated high-throughput technology and deep learning algorithms, etc., and for the first time realize the selectively hindered metastation aromatic C–H bond activation, and establish a well-performing yield prediction model, which provides a new paradigm for the combination of organic chemistry and artificial intelligence. Liao Minbiao and Yang Yuedong are the corresponding authors of the paper; Qiu Jia and Xie Jiancong are co-first authors.

As one of the research hotspots of organic chemistry, there have been a lot of selective aryl C–H bond activation work in recent years (Figure 1A). However, much of the reported progress has focused on the activation of adjacent, paranormal, and unhindered intermediate C–H bonds, while the activation of blocked interstitial C–H bonds has been hardly pursued. In order to solve such a challenging problem, the Liao Minbiao research group in the Guangzhou laboratory implemented a “three-step, one-pot” hindered m-amp;B C–H bond activation reaction (Figure 1B). These include a photoinitiated benzyl C–H bond carboxylation, carboxyl-directed Pd-catalyzed C–H bond activation, and finally microwave-induced decarboxylation. In the process of substrate expansion, automated high-throughput experimental techniques were used to screen a total of 24 ketones and 43 aryl potassium trifluoroborate, for a total of 1032 reactions. Notably, more than 70% of these reactions were targeted, demonstrating the ubiquity of the strategy. However, the complex combination of substrates makes the factors affecting yield very diverse. In fact, predicting the yield of unknown substrates can be tricky even for experienced chemists. Therefore, a predictive model needs to be built to lower the threshold for the use of this methodology and make it more practical. Finally, through cooperation with Yang Yuedong’s research group at Sun Yat-sen University, a communicative message passing neural network for reaction yield prediction (CMPRY) was jointly developed by combining organic chemistry with deep learning algorithms. With 5-fold cross validation (5cv), CMPRY can predict R2 = 0.75 and MAE = 7.2%. And the prediction for the test set remained good, at R2 = 0.73 and MAE = 6.6%. This result shows that the model has good epitaxiality and stability.

Figure 1: Regionally selective C–H activation strategy for replacing aromatic hydrocarbons. A. Guided aromatic C–H activation strategies; B. Hindered meta-position sp2 C–H activation strategy facilitated by automation and deep learning.

Taken together, the study used automation and machine learning to achieve a class of obstructed m-position C–H bond functionalization responses. Such a tandem reaction involves photocatalytic C–H bond carboxylation, Pd-catalyzed C–H bond arylation, and microwave-induced decarboxylation. Such an automated high-throughput technology combined with deep learning algorithms to explore the mode of hindered meta-position C–H activation hopes to promote a better combination of organic chemistry and artificial intelligence, and inject new impetus into the development of traditional basic disciplines.

The research work has been supported by scientific research funds from institutions and projects such as the National Natural Science Foundation of China and the Guangzhou Laboratory. (Source: Science Network)

Related paper information:

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button